Hypothetical and Real Consumer Choices Differentially Activate Common Frontostriatal Brain Circuitry
3.5 Discussion and Conclusions
Our study was motivated by the fact that hypothetical statements of intended choice are often different from real choices. In commercial marketing research, for example, self- reported intentions to buy goods are one of the most widely used measures to forecast sales of existing products, to test and plan the launch of new products (Chandon et al.
2004; Infosino 1986; Jamieson & Bass 1989; Silk & Urban 1978; Urban et al. 1983), and to assess the effectiveness of marketing programs (Raghubir & Greenleaf 2006; Schlosser et al. 2006). However, the fact that so many new products test well and then sell poorly suggests that forecasts based on reported intentions are imperfect. In the public policy arena and markets for public goods (e.g., clean air, foreign policy), data on what people actually choose are limited. There are no special markets for these goods, so voters’
preferences for them are often coarsely inferred from reactions to political positions.
Therefore, public economists and health economists often try to measure actual preferences from hypothetical responses to survey questions about possible actions (Carson et al. 1996; Diamond & Hausman 1994; Mortimer & Segal 2008). In political science, polls provide valuable information about voters’ perceptions and preferences (Crespi 1989), but pre-election questions about voting intentions are always hypothetical.
The newest research treats survey responses as deliberate products of a psychological and economic process by respondents, in order to understand their likely biases better and improve surveys (McFadden et al. 2005).
Furthermore, even actual choices often have hypothetical future choices embedded in them. For example, a student might make a real choice of a college to attend, planning to major in pre-medical studies. The planned major is an intended or a hypothetical future choice because it can be easily reversed.
The major behavioral result of our study is that subjects did exhibit a tendency to say
“Yes” to questions about purchasing consumer goods more often in hypothetical choices than in real choices. This “yes bias” is present when the real choices are different goods and when they are the same goods, and is unaffected by counterbalancing the condition
order (see Section 3.6). The bias can be characterized parametrically by the hypothesis that an initially stated WTP (elicited before the fMRI scanning) is reduced during real choice by a multiplier θR, which is typically less than one (a median of .60 in these data).
Intuitively, subjects change their minds and make real choices as if they will pay about 40% less than they initially said they might. When the initial WTP is adjusted to fit actual choices, subjects’ choice frequencies then appear to be a smooth logistic function of the modified decision value, thati is, R WTPprice.
FMRI compares differential activity in response to mDV during presentation of the product image—the time at which we assume valuation occurs—in real trials and in hypothetical trials. Three ROIs emerge in this comparison: mOFC, VStr, and ACC.
Regions active in hypothetical choice in mOFC and VStr are subsets of the regions active in real choice. ACC is highly significantly active in response to mDV only in real choice.
Medial OFC activity is correlated with decision value in both hypothetical and real decision making. Studies have found that the OFC, especially the medial part, is responsible for encoding the reward value of food, pleasant smells, attractive faces, and abstract rewards such as money or avoiding an aversive outcome (Anderson et al. 2003;
Gottfried et al. 2003; Kim et al. 2006; O'Doherty et al. 2001; O’Doherty et al. 2003;
Small et al. 2001). Our study uses consumer objects that have many product features and therefore require substantial abstract processing (including integration of the monetary price) which expands the scope of rewards that the mOFC appears to encode.
The medial OFC region identified in this study is similar to areas found to encode economic decision values in previous studies (Hare et al. 2008; Knutson et al. 2007;
Plassmann et al. 2007). Interestingly, activity in mOFC is stronger (more sensitive to decision value) and also more spatially widespread in real choice than in hypothetical choice. It is quite possible that the hypothetical bias might be related to an enhanced value signal in the brain, but the heightened mOFC activity during real choice casts doubt on this possibility.
The VStr activation in response to mDV could also be encoding decision value, as shown in many studies (Hariri et al. 2006; Kable & Glimpcher 2007; McClure et al. 2007;
Yacubian et al. 2006). Or it could be encoding a prediction error—the deviation of a new reward from a predicted reward (in our case, the mDV of a good compared to recent mDVs), as suggested by a recent study which separates values and prediction errors (Hare et al. 2008). If a stronger VStr response reflects reaction to prediction error, then learning about values could be more rapid when choices are real, which is a new implication of these results that could be tested in future studies.
An important open question is what neural activity is actually creating the “yes bias.”
Linking the stated object WTP dollar values to later choices suggests the bias is due to overstatement of object values in the pre-scanning phase. This overstatement may be generated by preference uncertainty or ignoring other alternatives and opportunity costs—that is, evaluating an option in isolation from others in an ideal world (Tanner &
Carlson 2009). Our data show that in hypothetical choices, those overstated values seem to be used (since the discounting factor θH is around 1), but in real choices they are adjusted downward by the discount factor θR. This discount factor θR is .60 in the fMRI study and is comparably less than one, .40 in a separate behavioral study.
A candidate region for implementation of this adjustment of stated WTP is the ACC. The ACC is only activated strongly, in response to mDV, in real decision making.
Furthermore, functional connectivity between the ACC and the mOFC is stronger in real trials as compared to hypothetical trials. In addition, in real choice, the probabilities of purchase are more sensitive to modified value and response times are faster (for “Yes”
decisions); these are possible behavioral manifestations of additional activity in ACC and mOFC. These results are consistent with the hypothesis that ACC is implementing a stronger reduction of casually-expressed WTP, to create an adjusted mDV that guides decisions, when real money is on the line. This possibility is consistent with the established role of ACC in executive function, cognitive control, and Stroop tasks which require effortful overrides of highly learned automatic responses (Botvinick et al. 1999;
van Veen et al. 2001; Yeung et al. 2006). While we have not found any cross-subject differences in ACC activity which are related to the degree of WTP adjustment θR, studies aimed more closely at discovering such relations are promising.
One potential way to use fMRI and other measures is to classify trial-by-trial hypothetical
“Yes” choices into those which are likely to switch to real “No” choices and those which are likely to stick as real “Yes” choices. Economists have tested some simple behavioral adjustments to estimate the overall switch minus stick rate, but there is no ideal technique that works reliably (Cummings & Taylor 1999; Knoepfle et al. 2009; List 2001). Much as in lie detection (Ben-Shakhar & Elaad 2003; Davatzikosa et al. 2005; Gamer et al. 2007), reading hidden intentions (Haynes et al. 2007) and other domains (Haxby et al. 2001;
Mitchell et al. 2008), it is possible that neural activation could provide another basis for some kind of low-level “mind-reading” of this sort. Such an application could
conceivably improve high-stakes forecasts for political polling, product design, and personal commitment substantially. Unfortunately, there are not enough switches in our data to permit a powerful classification based on neural activity of switches versus sticks (table A3.3) as in Grosenick et al. (Grosenick et al. 2008). However, a simple classification based on mDV and RTs predicts hypothetical-to-real switches and sticks with about 70% out-of-sample accuracy compared to a normalized 50% baseline—
namely, when mDV is low and responses are fast, hypothetical “Yes” decisions are likely to turn into real “No” decisions. (table A3.4). This baseline might be improved by more targeted studies using fMRI and other methods.
Finally, our study has an important methodological implication for scientific practice. In many experiments, especially in psychology and neuroscience, it is common to elicit hypothetical choices or ask hypothetical questions which cannot be actually implemented for practical reasons (e.g., for very high stakes, payments with long delays, creating unusual highly controlled social events, or morally charged consequences (Delgado et al.
2005; Greene et al. 2004; Greene et al. 2001; Hariri et al. 2006; Monterosso et al. 2007;
Takahashi et al. 2009)).
Generalizing claims about neural processing based on hypothetical choices to real choice assumes that the neural processes in hypothetical and real choice are highly overlapping.
Fortunately, our study shows that this overlap is largely present, at least in the narrow domain of consumer goods purchase. That is, suppose our study had used only hypothetical choices, concluded that mOFC and VStr are encoding decision value of consumer goods, and then guessed that the same regions would be active in response to valuation during real choices of consumer goods. Our results show that this guess would
have been correct (while understating signal strength and spatial extent). However, such a study would have failed to show that ACC responds to decision value, and might have concluded, incorrectly, that ACC plays no role in real choice.
As we have noted, for many phenomena in natural and social sciences, collecting hypothetical choice data is all we can do, even though the goal of collecting those data is to understand and predict real choices. Further studies should therefore continue to explore both similarities and differences in hypothetical and real choices, in different choice domains and with an eye to interesting applications.